Language selection contributes to intrusion errors in speaking: Evidence from picture naming
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Bilingualism: Language and Language selection contributes to intrusion
Cognition
errors in speaking: Evidence from
cambridge.org/bil picture naming
Xiaochen Zheng1 , Ardi Roelofs2 and Kristin Lemhöfer2
Research Article 1
Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, The Netherlands
International Max Planck Research School for Language Sciences, Max Planck Institute for Psycholinguistics,
Cite this article: Zheng X, Roelofs A, Lemhöfer Nijmegen, The Netherlands and 2Donders Institute for Brain, Cognition and Behaviour, Radboud University,
K (2020). Language selection contributes to
Nijmegen, The Netherlands
intrusion errors in speaking: Evidence from
picture naming. Bilingualism: Language and
Cognition 23, 788–800. https://doi.org/10.1017/ Abstract
S1366728919000683
Bilinguals usually select the right language to speak for the particular context they are in, but
Received: 4 February 2019 sometimes the nontarget language intrudes. Despite a large body of research into language
Revised: 26 August 2019 selection and language control, it remains unclear where intrusion errors originate from.
Accepted: 6 September 2019 These errors may be due to incorrect selection of the nontarget language at the conceptual
First published online: 31 October 2019
level, or be a consequence of erroneous word selection (despite correct language selection)
Keywords: at the lexical level. We examined the former possibility in two language switching experiments
bilingual; language intrusion; language using a manipulation that supposedly affects language selection on the conceptual level,
selection; cross-language interference namely whether the conversational language context was associated with the target language
Author for correspondence:
(congruent) or with the alternative language (incongruent) on a trial. Both experiments
Xiaochen Zheng, showed that language intrusion errors occurred more often in incongruent than in congruent
E-mail: x.zheng@donders.ru.nl contexts, providing converging evidence that language selection during concept preparation is
one driving force behind language intrusion.
Introduction
Most of the time, bilingual speakers succeed in selecting their target language for speaking in a
given language context and avoiding interference from a nontarget language (Poulisse, 1999;
Poulisse & Bongaerts, 1994). Only occasionally LANGUAGE INTRUSION ERRORS occur, which con-
cern the involuntary use of words from the nontarget language, such as a Dutch–English bilin-
gual saying “where is my fiets” to her English-speaking friend when she finds her bike stolen
(“fiets” is the Dutch word for “bike”). Such errors may happen in different contexts, for
example, after a change of interlocutor or in the presence of interfering background conversa-
tion. The rarity of language intrusion errors suggests strong language control mechanisms that
normally keep the languages apart (Declerck & Philipp, 2015; Green, 1998; Green & Wei,
2014). The mechanisms underlying language control and language selection have been studied
extensively: for example, using picture-word interference and language switching paradigms
(e.g., Boukadi, Davies & Wilson, 2015; Hermans, Bongaerts, De Bot & Schreuder, 1998;
Meuter & Allport, 1999; Zheng, Roelofs, Farquhar & Lemhöfer, 2018a). However, it has
remained unclear why and where in the speaking process language intrusion errors, as a failure
of control over target language production, may take place.
Producing a spoken word requires first preparing the intended concept to be expressed, and
then continuing to generate the word through lexical selection (e.g., Levelt, 1993; Levelt, Roelofs
& Meyer, 1999). According to several models of bilingual word production, the intention to use
one language rather than another is specified at the conceptual level (e.g., De Bot, 2004; La Heij,
2005; Roelofs, 1998; Roelofs, Dijkstra & Gerakaki, 2013; see also Green, 1998), which then fur-
ther drives the language-specific planning processes, including the selection of the words at the
lexical level in the appropriate language. For example, correctly naming a picture of a bike in
English by a Dutch–English bilingual speaker involves selection of the target language (i.e.,
English) at the conceptual level, followed by the planning of the English word bike at the lexical
© The Author(s) 2019. This is an Open Access level. The intrusion error “fiets” may occur because the speaker erroneously selected Dutch as the
article, distributed under the terms of the
target language at the conceptual level and then correctly planned the picture name in that lan-
Creative Commons Attribution licence (http://
creativecommons.org/licenses/by/4.0/), which guage. Alternatively, the intrusion may happen when English was correctly selected as the target
permits unrestricted re-use, distribution, and language, but at the lexical level, the Dutch word fiets was nevertheless incorrectly selected. The
reproduction in any medium, provided the latter may occur because both languages are still activated regardless of a bilingual’s intention to
original work is properly cited. speak one language only (Colomé, 2001; Costa, Miozzo & Caramazza, 1999; Hermans et al.,
1998). The current study aims at shedding more light on the question of where in the speaking
process language intrusion errors can originate from.
Cross-language interference is typically observed in bilingual picture-word interference
studies (e.g., when so-called “phono-translation” distractors are used). In the task, participants
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683
Bilingualism: Language and Cognition 789
are asked to name pictures in a given language (e.g., name the pic- been used in the previous trials. Thus, any language intrusion
ture mountain in English) while ignoring visual or auditory words errors in this situation are likely the consequence of incorrect lan-
in the same or the alternative language. When distractors are guage selection at the conceptual level, while language switch
words from the nontarget language (e.g., a Dutch word berm) errors can result from both lexical and conceptual-level interfer-
that phonologically overlap with the picture name in the nontar- ence. Being able to study this kind of intrusion error would
get language (e.g., berg, the Dutch word for mountain), they slow thus help us to isolate conceptual language intrusion errors
down naming response time (RT) and increase error rates (the from those arising from lexical processing.
so-called “phono-translation effect”). The interference is not It is worth noting that language intrusions have also been
only observed for distractors from the more dominant first lan- investigated extensively using a reading aloud task, where partici-
guage (L1) during naming in the less dominant second language pants are asked to read aloud mixed-language paragraphs (Gollan
(L2) (Boukadi et al., 2015; Hermans et al., 1998), but also the & Goldrick, 2018; Gollan, Schotter, Gomez, Murillo & Rayner,
other way around (Klaus, Lemhöfer & Schriefers, 2018). In 2014; Li & Gollan, 2018; Schotter, Li & Gollan, 2019). However,
these picture-word interference studies with phono-translation the fact that people can read aloud non-existing words suggests
distractors, intrusion errors are occasionally observed (in the cur- that reading aloud does not necessarily involve concept and
rent example, saying the Dutch word berg instead of the target lemma selection. Therefore, we consider the literature on reading
English word mountain), although not frequently. These intru- aloud to be less relevant for answering the current research ques-
sion errors can occur due to the incorrect selection during either tion and keep a discussion of it for later.
concept preparation or lexical selection. For example, because the As discussed so far, language intrusion takes place in daily life
Dutch word berg was primed by the phonologically-related dis- – though not very frequently (Muysken, 2000; Poulisse, 1999) – as
tractor berm, it may be erroneously selected at the lexical level well as in laboratory experiments, such as in the picture-word
even though the target language (English) had been correctly interference task (Boukadi et al., 2015; Hermans et al., 1998;
selected at the concept level. Alternatively, it is also possible Klaus et al., 2018), the cued language-switching task (Meuter &
that it was the nontarget language Dutch as a whole that was Allport, 1999; Zheng et al., 2018a; Zheng et al., 2018b), and the
primed by the Dutch distractor word berm, and therefore the lan- reading aloud task (Gollan & Goldrick, 2018; Gollan et al.,
guage itself was erroneously selected for naming. 2014; Li & Gollan, 2018; Schotter et al., 2019). Studying why
Besides in bilingual picture-word interference studies, language intrusion errors happen can help us better understand how bilin-
intrusion errors are also observed in language switching studies, guals exert control over the bilingual word production system.
where bilingual speakers are asked to name pictures while switch- The aim of the present study was to examine whether language
ing between their languages according to a given cue (e.g., a flag intrusion errors can happen as a result of incorrect language
or a color patch in addition to the to-be-named picture). In such a selection during concept preparation, rather than as a result of
paradigm, intrusion errors happen mostly in trials where partici- cross-language interference which takes place during lexical selec-
pants are required to switch the language relative to the previous tion (we certainly do not exclude the latter as a possibility,
trial (Zheng, Roelofs & Lemhöfer, 2018b). For example, after although we do not investigate this here). To this end, we experi-
consecutively naming pictures in English (e.g., ant, spoon, key), mentally created laboratory paradigms inspired by real-life scen-
a Dutch–English bilingual speaker may fail to switch to Dutch arios where language intrusions are likely to happen due to
but continue to name the picture tree in English instead of priming of the nontarget language at the conceptual level rather
using the target Dutch name boom. The mechanism of such than the lexical level. For that, we manipulated the language con-
intrusion errors, or the failure to switch, is also unclear. It is pos- text, which presumably will affect language activation in bilingual
sible that the speaker fails to implement the language switch at the speech production (see Hartsuiker, 2015, for a review).
conceptual level and consequently selects the previous language We developed two versions of bilingual picture naming tasks.
(English). Alternatively, it is also possible that the new target lan- In the first experiment, we simulated the situation where the lan-
guage (Dutch) has been correctly selected, but the planning of the guage associated with the interlocutor is incoherent with the con-
word during lexical selection is interfered by previously-selected versational environment (e.g., when you always speak English at
words from the nontarget language (e.g., the English words key, school, but one day it becomes more difficult because your sister,
spoon, ant). Besides the failure to switch to another language, lan- with whom you always talk in Dutch, is also there). Bilingual par-
guage intrusion also occurs when failing to stay in the same lan- ticipants were cued to speak a given language in the context of a
guage. While this type of error is less frequent than switch errors cartoon interlocutor who was associated with the same language
in the laboratory switching paradigm (e.g., Declerck, Lemhöfer & (congruent) or the different language (incongruent) as the target
Grainger, 2017; Zheng et al., 2018a; Zheng et al., 2018b), it does at language associated with the “environment” (location of the
least occasionally happen in real life (e.g., accidentally producing a to-be-named picture on the screen). In the second experiment,
Dutch word in an English conversation). To our purposes, these we simulated the distraction of background noise during daily
intrusion errors may be better suited to understand the process of conversation (e.g., when you are talking with your English-
language selection than the failure to switch: when one should speaking friend in a bar, but everyone around you is speaking
stay in the same language but fails to do so in a given language Dutch). Bilingual participants were cued to speak in a given lan-
context (e.g., a change of interlocutor or interfering background guage to an interlocutor while listening to the same (congruent)
conversation), it is less likely that the interference comes from or the alternative language (incongruent) as distractors. In both
nontarget-language words at the lexical level. Even though both experiments, the contextual congruency manipulation concerned
the target word and its translation-equivalent are activated during the language (conceptual level) rather than words in the language
production (e.g., Declerck, Philipp & Koch, 2015; Green, 1998), (lexical level). Therefore, if language intrusion errors happened
such activation remains low on repeat trials, in which the same because the nontarget language is selected, then intrusion errors
language is required as in the previous trial, as compared to should be found more often in incongruent than in congruent
switch trials, because words in the nontarget language have not contexts. We embedded the tasks in a mixed-language situation
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683
790 Xiaochen Zheng, Ardi Roelofs and Kristin Lemhöfer
(i.e., language switching) in order to induce higher levels of (CELEX database; Baayen, Piepenbrock & Gulikers, 1995) in
general cross-language interference. To be able to investigate the both Dutch and English. We matched all the Dutch and
‘pure’ process of language selection (and possible failures), we English picture names as closely as possible on number of sylla-
focused the analysis only on the repeat trials. Besides the contrast bles ( p = .813) and phonological onset category (e.g., fricatives
between the congruent and incongruent conditions, we also like /f/ have a delayed voice-key onset compared to vowels like
expected to observe that the dominant language is more likely /a/). Based on a pilot study on naming agreement, we replaced
to be intruded by the nondominant language than vice versa, two out of the 40 original pictures with drawings sketched by
which would replicate the reverse-dominance phenomenon in a the first author (see Appendix A for the full set of stimuli). All
mixed-language context (e.g., Gollan & Goldrick, 2018; Gollan the pictures were edited to a size of 300 × 300 pixels.
et al., 2014; Schotter et al., 2019; Zheng et al., 2018b).
Design
Each experimental session consisted of 640 trials, divided into
Experiment 1
eight blocks of 80 trials. Each picture appeared twice in a block,
In the first experiment, we instructed participants to name pic- once in Dutch and once in English. Twenty-five percent of the
tures either in English or in Dutch, depending on the location trials were switch trials. We pseudo-randomized all the items in
on the screen where the target picture was presented (‘conversa- each block using the program MIX (van Casteren & Davis,
tional environment’, valid cues). The invalid cues were cartoon 2006), with the following requirements: (1) subsequent trials
characters presented next to the target picture. They were intro- were semantically and phonologically unrelated; (2) repetition
duced as either English- or Dutch-speaking interlocutors. The of a picture was separated by at least four intervening trials; (3)
invalid cues could be congruent (indicating the same language) there were no more than six subsequent trials in the same
or incongruent (indicating the alternative language) with the language; (4) there were no subsequent switch trials.
valid cues. After having learned the association between interlocu- We manipulated interlocutor-location congruency (congruent
tors and language, participants were asked to ignore the invalid vs. incongruent) in both languages (L1 vs. L2). Twenty-five per-
cues (interlocutor) and focus on the valid cues (location). cent of all the trials were incongruent trials, i.e., the language
Crucially, we had the incongruent cues only on repeat trials. required by the object location was not the same as the language
Therefore, if an intrusion error occurred, it was most likely to associated with the interlocutor. All the items were proportionally
be a result of the incorrect selection of the nontarget language distributed across congruency conditions (i.e., 25 percent of
itself (which was primed by the incongruent interlocutor) at the the time an item occurs on an incongruent trial). To avoid the
conceptual level, rather than the immediate cross-language inter- co-occurrence of incongruency and switching, incongruent trials
ference from the previous trial during word selection – unlike a only occurred as repeat trials. We also made sure that there
switch trial, the nontarget language had not been actively used were no subsequent incongruent trials within a list. A second
on the previous trials. To make the experiment more naturalistic, list was constructed by reversing the block order of the first list.
we introduced the cartoon interlocutor and the naming task as
part of a real-life scenario, as explained below. Procedure
Participants were seated in a sound-proof booth and the experi-
ment was run using the software package Presentation (Version
Method
17.0, Neurobehavioural System Inc, Berkeley, U.S.). The computer
Participants screen (Benq XL2420Z, screen size 24 inch) was set to grey, with a
Twenty-two participants took part in the experiment for course resolution of 1920 × 1080 pixels, at a refresh rate of 120 Hz.
credit or vouchers. All of them were native Dutch speakers, First, the participants were familiarized with all picture names:
raised monolingually, who spoke English as their most proficient they saw each picture and named it in Dutch (block 1) or English
nonnative language. All the participants had normal or (block 2). After each picture naming, they were told the correct
corrected-to-normal vision. Data from two participants were answer and asked to name it again in case the original answer
excluded because they misunderstood the task or did not follow had been incorrect.
the instructions, leaving a final set of 20 participants (eight After that, we introduced the participants to the two Dutch- and
males). Table 1 shows the language background and English English-speaking cartoon interlocutors. Both interlocutors were
vocabulary size (measured by the LexTALE test, Lemhöfer & introduced as bilinguals with one of their languages being strongly
Broersma, 2012) of the final set of participants of this experiment dominant. The participants named all the pictures either in Dutch
and the later reported Experiment 2. or in English, according to the interlocutor presented next to the
picture. To make the interlocutors more salient, we used a
Materials 100-pixel-wide color frame for the pictures when the corresponding
Experimental stimuli consisted of 40 black-and-white line interlocutor was presented (blue frame for the English-speaking
drawings, representing 40 translation pairs of Dutch–English interlocutor and orange frame for the Dutch-speaking interlocutor).
noncognate words (e.g., the Dutch word “boom” and its This served as a training of the interlocutor-language association.
English translation “tree”). All the pictures were selected from The correct word was presented on the screen after each response
the international picture naming project (IPNP) database for the first ten trials and then the training continued for another
(Bates, D’amico, Jacobsen, Székely, Andonova, Devescovi, 30 trials without feedback. Switch rate was kept the same as in
Herron, Lu, Pechmann, Pléh, Wicha, Federmeier, Gerdjikova, the main experiment (25%).
Gutierrez, Hung, Hsu, Iyer, Kohnert, Mehotcheva, Orozco- Then we introduced the participants to the location cues: the
Figueroa, Tzeng & Tzeng, 2003), opting for those with highest target pictures would be presented in one of the four corners of
naming agreements (Bates et al., 2003; Severens, Van Lommel, the screen, which represented either “school” or “home” (e.g.,
Ratinckx & Hartsuiker, 2005) and high lexical frequency top-left corner and bottom right corner for “school”, and top-
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683
Bilingualism: Language and Cognition 791
Table 1. Participants’ language background and English proficiency in Experiments 1 and 2.
Experiment 1 (N = 20) Experiment 2 (N = 29)
Characteristic Mean SD Range Mean SD Range
Age 22.2 2.7 18–26 21.4 2.4 18–31
Age of acquiring English 10.2 2.3 6–14 10.6 1.6 7–13
Self-rated frequency of using Englisha
- speaking 3.2 1.0 1–5 3.1 1.0 2–5
- listening 4.4 0.9 2–5 4.5 0.6 3–5
- writing 3.4 1.1 1–5 2.8 1.0 1–5
- reading 3.8 1.1 2–5 3.8 1.0 2–5
a
Self-rated frequency of switching languages
- speaking 2.2 0.9 1–4 2.1 0.9 1–4
a
Self-rated proficiency of English
- speaking 4.1 0.8 2–5 3.5 0.9 2–5
- listening 4.7 0.7 3–5 4.1 0.7 2–5
- writing 4.1 0.9 2–5 3.6 1.0 2–5
- reading 4.7 0.5 4–5 4.2 0.7 2–5
English vocabulary size
- LexTALE test 81.9 9.9 58–96b 72.7 9.8 48–93b
Note: SD = Standard Deviation.
a
Self-ratings were given on a scale from 1 = very rarely/bad to 5 = very often/good.
b
The score is a weighted % correct score, i.e., 50 is chance level, 100 is the maximum score.
right and bottom-left corner for “home”, or vice versa). At stimulus stayed on the screen for a total of 2550 ms. After another
“school” the participants were supposed to speak English whereas jittered blank screen of 250–500 ms, the next trial began. In total,
at “home” they spoke Dutch. Two locations were used to cue there were eight blocks of 80 trials. After each block, participants
each language, so that the location could alternate between each received feedback on their performance (e.g., speed) and got
trial to avoid a confound of language switch and location switch reminded of the languages represented by the locations. We
(Mayr & Kliegl, 2003). We counterbalanced the assignment of instructed them to name the pictures as quickly as possible in the
the locations to the response language across participants. After language indicated by the location cue, and also not to correct them-
ten trials, we introduced time pressure to induce more speech selves when they said something wrong. All the instructions were in
errors. For that, a response deadline was computed dynamically English.
and calibrated individually for each participant, based on the At the end of the session, the participants completed the
80th percentile of the previous ten trials. Participants would receive LexTALE vocabulary test in English (Lemhöfer & Broersma,
a warning message for being “too late” if they failed to respond 2012) and a language background questionnaire, as summarized
within the time limit. This continued for another 80 trials. in Table 1. The entire session took approximately 1.5 hrs.
During this phase, the interlocutor cues (that would become invalid
in the main experiment) were always congruent with the location Data analysis
cues (that would be the valid cues in the main experiment). Error rates and RTs were used as dependent variables. Only repeat
At the beginning of the main experiment, we introduced the trials were analyzed. Participants’ responses were coded either as
incongruent condition, i.e., when the interlocutor presented (1) correct, fluent responses, or as (2) incorrect responses.
next to the picture indicated a different language from the one Incorrect responses were further categorized into language intru-
indicated by the location cue (e.g., the participants would see sion errors (i.e., complete and fluent naming responses using the
the English-dominant interlocutor at “home”, where they were translation equivalent in the nontarget language) and eleven other
supposed to speak Dutch). Figure 1 shows a schematic diagram types of errors, such as self-corrections, disfluencies, or using a
for a trial where participants needed to name the picture in wrong word in the correct language. Correctly responded trials
English or in Dutch, in either the congruent or incongruent condi- with an RT (measured automatically by the voice key) deviating
tion. We instructed the participants to pay attention to the valid cues more than three standard deviations from the respective partici-
(i.e., locations). During the experimental blocks, each trial started pants’ condition mean were defined as another type of error
with the 250 ms presentation of a fixation cross, followed by a (i.e., RT outliers, see Appendix C for all the categories and the
blank screen with a jitter of 250–500 ms. Then, the picture appeared percentages of each type of error). We excluded all error trials
in one of the four corners of the screen, and the picture and the as well as post-error trials from the RT analysis. In the analysis
interlocutor stayed together on the screen until 550 ms after the of intrusion errors, we excluded trials at the beginning of each
voice key (Shure SM-57 microphone) had registered the onset of block and trials following language intrusion errors or other inter-
speech. If the voice key was not triggered within 2000 ms, the lingual errors (see Appendix C).
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683
792 Xiaochen Zheng, Ardi Roelofs and Kristin Lemhöfer
Fig. 1. A schematic diagram for Experiment 1. The target language was cued by the location (home: Dutch; school: English). Next to the picture, a cartoon inter-
locutor was simultaneously presented with a color frame, which was associated either with the target language (congruent condition, top panel) or the nontarget
language (incongruent condition, bottom panel). The diagram depicts an experimental trial where participants had to name the picture either in English (A) or in
Dutch (B).
We performed the statistical analyses using mixed-effects mod- Speakers made more language intrusion errors on incongruent
els with the lme4 package (Version 1.1.13, Bates, Mächler, Bolker & than on congruent repeat trials, and also more when naming in
Walker, 2015) in R (Version 3.4.1; R Core Team, 2017). The factors the L1 than in the L2. There was no interaction between language
language (L1 vs. L2) and congruency (congruent vs. incongruent) and congruency.
were sum-coded and included as fixed effects in the models. As for RTs, speakers were slower on incongruent than on
Participants and items were included as random effects. For both congruent trials and also slower when naming in the L1 than in
RT and error analyses, we used generalized linear mixed models the L2. There was a significant interaction between congruency
(GLMEMs). GLMEMs were chosen for the RT analysis to account and language. A follow-up analysis for each language showed that
for the right-skewed shape of the RT distribution without the need the congruency effect was larger in the L1 (ML1cong = 807 ms,
to transform and standardize the raw data (Lo & Andrews, 2015). SDL1cong = 63 ms; ML1incong = 883 ms, SDL1incong = 89 ms;
All the analyses were conducted with a maximal random-effects β = −39.61, SE = 6.93, t = −5.71, p < .001) than in the L2
structure, which includes random intercepts and random slopes (ML2cong = 754 ms, SDL2cong = 50 ms; ML2incong = 796 ms,
for all fixed effects and their interactions for both participants SDL2incong = 65 ms; β = −22.33, SE = 4.98, t = −4.49, p < .001).
and items (Barr, Levy, Scheepers & Tily, 2013). When the model To summarize, language intrusion errors were more likely
failed to converge, we simplified it by removing the interactions and responses were slower in the incongruent than in the con-
in the random structure (see Appendix D for the final models gruent contexts, and also in the dominant L1 than in the weaker
used for analyses). For both analyses, we reported Wald’s z-scores, L2. There was an interaction between congruency and language
t-scores and their associated p-values. dominance in the RTs: the congruency effect was larger in L1
than in L2. These findings suggest that language intrusion can
happen as a result of incorrect language selection on the concep-
Results tual level.
Speakers made different types of speech errors on 11.0% of all trials,
including language intrusion errors (e.g., said the Dutch word
Experiment 2
“boom” instead of the English word “tree” on an English trial)
on 4.4% of the repeat trials and 9.3% of the switch trials. In the second experiment, we sought for converging evidence for
Figure 2 shows the violin plots for the language intrusion error intrusion errors caused by incorrect language selection, using a
rates and the RTs on the repeat trials. Table 2 gives the statistics different paradigm inspired by real-life scenarios. Similar to
from the GLMEMs. Experiment 1, participants would see cartoon interlocutors and
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683Bilingualism: Language and Cognition 793
Fig. 2. Violin plots with individual data distributions for
language intrusion error rate (panel A) and mean RT
(panel B, in ms) on repeat trials, grouped by language
and congruency. The outer shapes represent the distri-
bution of individual data, the thick horizontal line inside
the box indicates the median, and the bottom and top
of the box indicate the first and third quartiles of each
condition.
Table 2. Statistics from the GLMEMs for language intrusion error rate (ER, in %) and reaction time (RT, in ms) on repeat trials in Experiment 1.
Mean (SD) β SE z- or t-value p
ER Language L1 (Dutch) 8.2 (6.8) 0.75 0.14 5.22794 Xiaochen Zheng, Ardi Roelofs and Kristin Lemhöfer
Fig. 3. A schematic diagram for Experiment 2. The target language was cued by the cartoon interlocutor with a color frame. Besides, an auditory distractor was
presented either in the target language (congruent condition, top panel) or in the nontarget language (incongruent condition, bottom panel [“boer” is the Dutch
translation of “farmer”]).The diagram depicts an experimental trial where participants had to name the picture either in English (A) or in Dutch (B) SOA: stimulus
onset asynchrony.
Materials Besides the restrictions used in Experiment 1, we made sure
The experimental materials consisted of 40 pictures and 40 pairs that the auditory distractors were semantically and phonologically
of auditory distractors which were translation equivalents between unrelated to the pictures after the current trial, to avoid potential
Dutch and English. The picture stimuli were identical to the ones priming effects.
used in Experiment 1.
Dutch and English auditory distractors were voice recordings of Procedure
a male Dutch native speaker. We did this to make sure that the The setup of the experiment was identical to Experiment 1, apart
accent of the audios was familiar enough to our participants. from that the computer screen was set to black instead of grey.
Auditory distractors were highly frequent words representing daily The testing procedure was similar. We describe it below, mainly
objects. The distractors were selected to be noncognate words focusing on its differences from Experiment 1.
between Dutch and English and were always presented with the First, the participants were familiarized with all picture names
same picture. Furthermore, auditory distractors were semantically and introduced to the two Dutch- and English-speaking interlo-
and phonologically unrelated to the target picture name in both lan- cutors. In Experiment 2, we used two same-gender interlocutors
guages (e.g., the target picture of “tree”, or “boom” in Dutch, was (i.e., two males) instead of the two different-gender interlocutors
presented with the English word “dust” or its Dutch translation used in Experiment 1, in order to be consistent with the same-
“stof” as auditory distractors). Lastly, syllable length of the target gender auditory distractors. Both interlocutors were introduced
picture name and the incongruent (i.e., other language) auditory as monolingual speakers. Unlike their distractor roles in
distractor was matched (see Appendix A for the full set of stimuli). Experiment 1, the interlocutors in Experiment 2 served as valid
cues. The interlocutors were presented together with a color
Design frame (blue frame for the English-speaking interlocutor and red
The design was identical to that in Experiment 1, with the inde- frame for the Dutch-speaking interlocutor) to make the primes
pendent variables being language (L1 vs. L2) and congruency more salient. Again, participants received 40 trials of training
(congruent vs. incongruent), and the dependent variables being for the interlocutor-language association, where they named the
intrusion error rates and RTs. The lists were constructed in a pictures either in English or in Dutch according to the interlocu-
similar way as in Experiment 1, pseudo-randomized by the pro- tor cues. After that, we asked the participants to put on head-
gram MIX. Twenty-five percent of the trials were switch trials phones and to name the pictures while at the same time being
and one third of the repeat trials were incongruent trials. presented with the auditory distractors. They were asked to
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683Bilingualism: Language and Cognition 795
Fig. 4. Violin plots with individual data distributions for
language intrusion error rate (panel A) and mean RT
(panel B, in ms) on repeat trials, grouped by language
and congruency. The outer shapes represent the distri-
bution of individual data, the thick horizontal line inside
the box indicates the median, and the bottom and top
of the box indicate the first and third quartiles of each
condition.
Table 3. Statistics from the GLMEMs for language intrusion error rate (ER, in %) and reaction time (RT, in ms) on repeat trials in Experiment 2.
Mean (SD) β SE z- or t-value p
ER Language L1 (Dutch) 6.9 (7.9) 0.54 0.10 5.21796 Xiaochen Zheng, Ardi Roelofs and Kristin Lemhöfer
well as a language background questionnaire. The entire session to switch but fail to do so, which was the predominant line of
took approximately 1.5 hrs. inquiry in previous research (e.g., Meuter & Allport, 1999;
Zheng et al., 2018b). In both experiments, we observed more lan-
Data analysis guage intrusion errors and longer RTs on incongruent repeat
We used the same analysis procedures as in Experiment 1. trials (i.e., when the interlocutor and location cues were not in-
dicating the same language in Experiment 1; or when the back-
ground and current ‘conversation’ were not in the same
Results language in Experiment 2) than on congruent trials. In both
Speakers made different types of speech errors on 9.6% of all cases, the congruency manipulation concerned language selection
trials, including language intrusion errors on 4.1% of the repeat rather than word selection. Therefore, language intrusion errors
trials and 8.0% of the switch trials. that were due to incongruency can be attributed to the erroneous
Figure 4 shows the violin plots for language intrusion error selection of the nontarget language.
rates and RTs on the repeat trials. Table 3 gives the statistics In Experiment 1, we associated the cartoon characters (the
from the GLMEMs. invalid cues) with one of the two languages, rather than with
Speakers made more language intrusion errors on incongruent any specific words in the languages. Therefore, the intrusion
than on congruent repeat trials and when naming in the L1 than errors caused by the congruency manipulation (i.e., more intru-
in the L2. There was no significant interaction between language sion errors were observed when the interlocutor was associated
and congruency. with the nontarget language) were likely to be a result of the non-
As for RTs, speakers were slower on incongruent than on target language being primed. Interference on the lexical level due
congruent trials and when naming in the L1 than in the L2. to the congruency manipulation is unlikely, otherwise the incon-
There was also a significant interaction between congruency and gruent interlocutor would have to boost the activation of the
language. A follow-up analysis for each language showed that whole lexicon in the nontarget language. Note that although the
the congruency effect was only present in the L2 (ML2cong = 732 results of the RTs showed the same pattern as the errors (i.e.,
ms, SDL2cong = 61 ms; ML2incong = 762 ms, SDL2incong = 68 ms; longer RTs in the incongruent than in the congruent condition),
β = −15.49, SE = 3.03, t = −5.12, p < .001), but not in the L1 this is not direct evidence for incorrect language selection because
(ML1cong = 835 ms, SDL1cong = 89 ms; ML1incong = 842 ms, these RTs were obtained in correct trials. Nevertheless, the pro-
SDL1incong = 83 ms; β = −3.16, SE = 3.77, t = −0.84, p = .402). longed RTs may reflect the difficulty in selection which resulted
To summarize, language intrusion errors were more likely and from additional activation of the competitive language. Our results
correct responses were slower in the incongruent than in the con- are coherent with the idea that language context, such as faces
gruent contexts. Besides, responses were slower and less accurate associated with a certain social-cultural identity, affects language
in the dominant L1 than in the weaker L2. There was an inter- production (e.g., Blanco-Elorrieta & Pylkkänen, 2017;
action between congruency and language dominance in the Hartsuiker, 2015; Li, Yang, Scherf & Li, 2013; Liu, Timmer, Jiao,
RTs: the congruency effect was only present in the L2. Despite Yuan & Wang, 2019).
the interaction in RTs (in which the congruency effect was larger In Experiment 2, we used distractor words (e.g., “stof” or its
in the L1 than in the L2), these findings converge with those of English translation “dust”) that were unrelated to either the target
Experiment 1, suggesting that language intrusion can happen words (e.g., tree) or its translation equivalent (the Dutch word
due to the incorrect selection of language during concept boom). In the incongruent condition, these distractor words
preparation. were from the nontarget language, while congruent distractors
were from the target language. Therefore, the occurrence of
more intrusion errors in the incongruent than in the congruent
General discussion condition is again more likely to be a result of the selection of
the nontarget language (in this case, Dutch) which was primed
Inferences about bilingual control mechanisms can be made by by the distractor word. The errors are unlikely to be due to cross-
studying how and when these mechanisms fail, e.g., when lan- language interference during word selection, which has been
guage intrusions occur. In the current study, we examined observed when distractor words have a specific relation to the tar-
whether language intrusion errors may be the result of selecting get words like in the phono-translation condition in the picture-
the nontarget language itself at the conceptual level rather than word interference task (e.g., the distractor word berm priming the
selecting a word from the nontarget language at the lexical level Dutch word berg; Hermans et al., 1998). Although not particu-
(while the language has been correctly selected). In the first larly investigated, the fact that merely listening to the nontarget
experiment, we introduced incongruent interlocutor-location pairs language could affect target language production is also consistent
(e.g., an English-Dutch bilingual interlocutor with English as the with the idea that language control mechanisms are shared
dominant language vs. the house of a Dutch-speaking family) in a between comprehension and production, and that bottom-up lin-
language switching task. In the second experiment, we combined guistic representations have a considerable influence on language
the language switching task with an auditory picture-word inter- selection processes in both modalities (Gambi & Hartsuiker, 2016;
ference task, to simulate the situation where background conver- Peeters, Runnqvist, Bertrand & Grainger, 2014).
sation is disturbing the selection of the target language (e.g., when It is also worth noting that intrusion errors that happened
the background conversation is in English whereas the current in the congruent condition, or that occurred on switch trials,
target language is Dutch). might still be attributable to erroneous lexical selection.
Although embedded in mixed-language contexts, we only Therefore, we do not reject the alternative possibility that cross-
looked at situations where the bilingual participants were sup- language interference during word selection can also lead to lan-
posed to stay in the same language (i.e., repeat trials) but failed guage intrusions. Actually, this type of intrusion is likely because
to do so – in contrast to situations where participants are asked both languages are activated regardless of a bilingual’s intention
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683Bilingualism: Language and Cognition 797
to speak one language only (Colomé, 2001; Costa et al., 1999; from the ‘staying in the same language’ situation in daily life.
Hermans et al., 1998). Nevertheless, our two experiments provide Future research can aim to find ways of inducing sufficient num-
converging evidence that incorrect language selection on the con- bers of intrusion errors in a monolingual mode.
ceptual level is one factor contributing to language intrusion To summarize, the current study investigated whether
errors. Moreover, although we interpret the incorrect language language intrusion errors can be caused by the erroneous selec-
selection on the conceptual level as a failure of control, it has to tion of the language on the conceptual level. We examined this
be acknowledged that language control goes beyond language in two experiments by manipulating language context: more spe-
selection and takes place at multiple levels of processing (e.g., cifically, the congruency of two language cues (one task-relevant,
Declerck & Philipp, 2017; Gollan et al., 2014; Olson, 2013). one non-relevant). In both experiments, we observed that lan-
In both experiments, language intrusion errors were more guage intrusion errors occurred more often when the context
likely and responses were slower in the dominant L1 than the was incongruent than congruent with the target language. This
weaker L2. This finding replicates the so-called REVERSE finding provides evidence that language selection, rather than
DOMINANCE EFFECT, i.e., the seemingly paradoxical finding that pro- only selection at the lexical level, is an error-prone process during
duction in the dominant L1 can under some circumstances be bilingual word production.
more difficult than in the L2. This effect is reliably observed in
standard cued language switching experiments (Christoffels, Data availability. Data are available from the Donders Institute for Brain,
Cognition and Behaviour repository at http://hdl.handle.net/11633/aab2nrxz.
Firk & Schiller, 2007; Costa & Santesteban, 2004; Verhoef,
Roelofs & Chwilla, 2009; Zheng et al., 2018b), and has also been Acknowledgements. The authors would like to thank Kristin Witte, Max
shown for voluntary language switching (Gollan & Ferreira, Sterling, Daniel Holthuis, Elena Nicolaou, and Rick Alofs for their help in
2009) and for other language-mixing tasks (Gollan & Goldrick, data collection.
2018; Gollan et al., 2014; Schotter et al., 2019). This reverse dom-
inance effect can be accounted for by assuming that when unba- References
lanced bilinguals mix languages, they need to inhibit the
dominant language while enhancing the less dominant language Allport A and Wylie G (1999) Task-switching: Positive and negative priming
to facilitate production (Allport & Wylie, 1999; Green, 1998). of task set. In Humphreys J, Duncan J and Treisman A (Eds.), Attention,
Additionally, we found an interaction between the congruency space, and action: Studies in cognitive neuroscience. Oxford: Oxford
University Press, pp. 273–296.
effect and language dominance in both experiments. However, the
Baayen H, Piepenbrock R and Gulikers L (1995) The CELEX Lexical
direction of this interaction differed between experiments. In the Database [CD-ROM]. Linguistic Data Consortium, University of
first experiment, the congruency effect was larger in the L1 than Pennsylvania, Philadelphia PA.
in the L2, whereas, in the second experiment, the congruency Barr DJ, Levy R, Scheepers C and Tily HJ (2013) Random effects structure
effect was restricted to the L2. The interaction was only observed for confirmatory hypothesis testing: Keep it maximal. Journal of Memory
in the RTs, but not in the intrusion error rates. The difference in and Language 68, 255–278. https://doi.org/10.1016/j.jml.2012.11.001
interaction might be due to the fact that we used different manip- Bates D, Mächler M, Bolker B and Walker S (2015) Fitting linear mixed-
ulations of contextual priming in the two experiments: in the first effects models using lme4. Journal of Statistical Software 67, 1–48. https://
experiment, the congruency between the interlocutor’s identity doi.org/10.18637/jss.v067.i01
(invalid cue) and picture location (valid cue) was manipulated; Bates E, D’amico S, Jacobsen T, Székely A, Andonova E, Devescovi A,
Herron D, Lu C, Pechmann T, Pléh C, Wicha N, Federmeier K,
in the second experiment, the distracting information came
Gerdjikova I, Gutierrez G, Hung D, Hsu J, Iyer G, Kohnert K,
from the language spoken in the background. Whereas face or cul- Mehotcheva T, Orozco-Figueroa A, Tzeng A and Tzeng O (2003)
ture priming can facilitate L1 and L2 picture naming (Li et al., Timed picture naming in seven languages. Psychonomic Bulletin &
2013), auditory distractor words may cause more interference Review 10, 344–380. https://doi.org/10.3758/BF03196494
rather than facilitation. Unfortunately, we did not include a neutral Blanco-Elorrieta E and Pylkkänen L (2017) Bilingual language switching in
condition in the experiments to be able to attribute the congruency the laboratory versus in the wild: The spatiotemporal dynamics of adaptive
effect to inhibition of the incongruent condition, to facilitation of language control. Journal of Neuroscience 37, 9022–9036. https://doi.org/10.
the congruent condition, or both. Therefore, the precise reason 1523/JNEUROSCI.0553-17.2017
for the difference in the direction of the interaction in the two Boukadi M, Davies R and Wilson MA (2015) Bilingual lexical selection as a
experiments remains unclear and requires further investigation. dynamic process: Evidence from Arabic-French bilinguals. Canadian
Journal of Experimental Psychology 69, 297–313. https://doi.org/10.1037/
In the current study, we also attempted to take a more eco-
cep0000063
logically valid approach to investigating naturally occurring lan- Christoffels IK, Firk C and Schiller NO (2007) Bilingual language control:
guage intrusions by employing two novel versions of a bilingual An event-related brain potential study. Brain Research 1147, 192–208.
switching paradigm. Compared to classic language-switching https://doi.org/10.1016/j.brainres.2007.01.137
tasks where participants are cued to switch (i.e., switch trials), Colomé À. (2001) Lexical activation in bilinguals’ speech production:
the current paradigms focus on repeated naming (i.e., repeat Language-specific or language-independent? Journal of Memory and
trials) and are able to look into more natural aspects of the failures Language 45, 721–736. https://doi.org/10.1006/jmla.2001.2793
of language selection. By manipulating the language context, the Costa A, Miozzo M and Caramazza A (1999) Lexical selection in bilinguals:
paradigms successfully simulated daily-life scenarios where lan- Do words in the bilingual’s two lexicons compete for selection? Journal of
guage intrusion is more likely to occur. Compared to other tasks Memory and Language 41, 365–397. https://doi.org/10.1006/jmla.1999.2651
Costa A and Santesteban M (2004) Lexical access in bilingual speech produc-
such as the reading aloud of texts (Gollan & Goldrick, 2018;
tion: Evidence from language switching in highly proficient bilinguals and
Gollan et al., 2014; Li & Gollan, 2018; Schotter et al., 2019), the cur- L2 learners. Journal of Memory and Language 50, 491–511. https://doi.org/
rent paradigm is better suited to investigate failures to ‘stay’ rather 10.1016/j.jml.2004.02.002
than failures to switch. However, in the current study, bilingual par- De Bot K (2004) The multilingual lexicon: Modelling selection and control.
ticipants were still asked to use both of their languages in quick suc- International Journal of Multilingualism 1, 17–32. https://doi.org/10.1080/
cession, which makes the repeat trials still intrinsically different 14790710408668176
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683798 Xiaochen Zheng, Ardi Roelofs and Kristin Lemhöfer
Declerck M, Koch I and Philipp AM (2015) The minimum requirements of Li C and Gollan TH (2018) Cognates interfere with language selection but
language control: Evidence from sequential predictability effects in language enhance monitoring in connected speech. Memory & Cognition 46, 923–
switching. Journal of Experimental Psychology: Learning, Memory, and 939. https://doi.org/10.3758/s13421-018-0812-x
Cognition 41, 377. http://doi.org/10.1037/xlm0000021 Li Y, Yang J, Scherf S and Li P (2013) Two faces, two languages: An fMRI
Declerck M, Lemhöfer K and Grainger J (2017) Bilingual language interfer- study of bilingual picture naming. Brain and Language 127, 452–462.
ence initiates error detection: Evidence from language intrusions. https://doi.org/10.1016/j.bandl.2013.09.005
Bilingualism: Language and Cognition 20, 1010–1016. https://doi.org/10. Liu C, Timmer K, Jiao L, Yuan Y and Wang R (2019) The influence of con-
1017/S1366728916000845 textual faces on bilingual language control. Quarterly Journal of
Declerck M and Philipp AM (2015) A review of control processes and their Experimental Psychology. https://doi.org/10.1177/1747021819836713
locus in language switching. Psychonomic Bulletin & Review 22, 1630–1645. Lo S and Andrews S (2015) To transform or not to transform: using general-
https://doi.org/10.3758/s13423-015-0836-1 ized linear mixed models to analyse reaction time data. Frontiers in
Declerck M and Philipp AM (2017) Is there lemma-based language control? Psychology 6, 1–16. https://doi.org/10.3389/fpsyg.2015.01171
The influence of language practice and language-specific item practice on Mayr U and Kliegl R (2003) Differential effects of cue changes and task
asymmetrical switch costs. Language, Cognition and Neuroscience 32, changes on task-set selection costs. Journal of Experimental Psychology:
488–493. https://doi.org/10.1080/23273798.2016.1250928 Learning, Memory, and Cognition 29, 362–372. https://doi.org/10.1037/
Gambi C and Hartsuiker RJ (2016) If you stay, it might be easier: Switch 0278-7393.29.3.362
costs from comprehension to production in a joint switching task. Meuter RFI and Allport A (1999) Bilingual language switching in naming :
Journal of Experimental Psychology: Learning, Memory, and Cognition 42, Asymmetrical costs of language selection. Journal of Memory and
608. http://doi.org/10.1037/xlm0000190 Language 40, 25–40. https://doi.org/10.1006/jmla.1998.2602
Gollan TH and Ferreira VS (2009) Should I stay or should I switch? A cost– Muysken P (2000) Bilingual speech: A typology of code-mixing. Cambridge:
benefit analysis of voluntary language switching in young and aging bilin- Cambridge University Press.
guals. Journal of Experimental Psychology: Learning, Memory, and Olson DJ (2013) Bilingual language switching and selection at the phonetic
Cognition 35, 640–665. https://doi.org/10.1037/a0014981 level: Asymmetrical transfer in VOT production. Journal of Phonetics 41,
Gollan TH and Goldrick M (2018) A switch is not a switch: 407–420. https://doi.org/10.1016/j.wocn.2013.07.005
Syntactically-driven bilingual language control. Journal of Experimental Peeters D, Runnqvist E, Bertrand D and Grainger J (2014) Asymmetrical
Psychology: Learning, Memory, and Cognition 44, 143–156. https://doi. switch costs in bilingual language production induced by reading words.
org/10.1037/xlm0000462 Journal of Experimental Psychology: Learning, Memory, and Cognition 40,
Gollan TH, Schotter ER, Gomez J, Murillo M and Rayner K (2014) Multiple 284. http://doi.org/10.1037/a0034060
levels of bilingual language control: Evidence from language intrusions in Poulisse N (1999) Slips of the tongue: Speech errors in first and second language
reading aloud. Psychological Science 25, 585–595. https://doi.org/10.1177/ production. Philadelphia, PA: John Benjamins Publishing.
0956797613512661 Poulisse N and Bongaerts T (1994) First language use in second language
Green DW (1998) Mental control of the bilingual lexico-semantic system. production. Applied Linguistics 15, 36–57. https://doi.org/10.1093/applin/
Bilingualism: Language and Cognition 1, 67–81. https://doi.org/10.1017/ 15.1.36
S1366728998000133 Roelofs A (1998) Lemma selection without inhibition of languages in bilingual
Green DW and Wei L (2014) A control process model of code-switching. speakers. Bilingualism: Language and Cognition 1, 94–95. https://doi.org/10.
Language, Cognition and Neuroscience 29, 499–511. https://doi.org/10. 1017/S1366728998000194
1080/23273798.2014.882515 Roelofs A, Dijkstra T and Gerakaki S (2013) Modeling of word translation:
Hartsuiker RJ (2015) Visual Cues for Language Selection in Bilinguals. In Activation flow from concepts to lexical items. Bilingualism: Language
R Mishra, N Srinivasan, F Huettig (Eds), Attention and Vision in and Cognition 16, 343–353. https://doi.org/10.1017/S1366728912000612
Language Processing. New Delhi: Springer https://doi.org/10.1007/978-81- Schotter ER, Li C and Gollan TH (2019) What reading aloud reveals
322-2443-3_8 about speaking: Regressive saccades implicate a failure to monitor, not
Hermans D, Bongaerts T, De Bot K and Schreuder R (1998) Producing inattention, in the prevalence of intrusion errors on function words.
words in a foreign language: Can speakers prevent interference from their Quarterly Journal of Experimental Psychology, https://doi.org/10.1177/
first language? Bilingualism: Language and Cognition 1, 213–229. https:// 1747021818819480.
doi.org/10.1017/S1366728998000364 Severens E, Van Lommel S, Ratinckx E and Hartsuiker RJ (2005) Timed
Klaus J, Lemhöfer K and Schriefers H (2018) The second language interferes picture naming norms for 590 pictures in Dutch. Acta Psychologica 119,
with picture naming in the first language: evidence for L2 activation during 159–187. https://doi.org/10.1016/j.actpsy.2005.01.002
L1 production. Language, Cognition and Neuroscience 3798. https://doi.org/ van Casteren M and Davis MH (2006) Mix, a program for pseudorandomiza-
10.17605/OSF.IO/HMDAX tion. Behavior Research Methods 38, 584–589. https://doi.org/10.3758/
La Heij W (2005) Selection processes in monolingual and bilingual lexical access. BF03193889
In J Kroll & A de Groot (Eds.), Handbook of bilingualism: Psycholinguistic Verhoef KMW, Roelofs A and Chwilla DJ (2009) Role of inhibition in
approaches. Oxford, England: Oxford University Press, pp. 289–307. language switching: Evidence from event-related brain potentials in overt
Lemhöfer K and Broersma M (2012) Introducing LexTALE: a quick and valid picture naming. Cognition 110, 84–99. https://doi.org/10.1016/j.cognition.
lexical test for advanced learners of English. Behavior Research Methods 44, 2008.10.013
325–343. https://doi.org/10.3758/s13428-011-0146-0 Zheng X, Roelofs A, Farquhar J and Lemhöfer K (2018a) Monitoring of lan-
Levelt WJM (1993) Speaking: From intention to articulation. Cambridge, MA: guage selection errors in switching: Not all about conflict. PLOS ONE 13,
MIT press. e0200397. https://doi.org/10.1371/journal.pone.0200397
Levelt WJM, Roelofs A and Meyer AS (1999) A theory of lexical access in Zheng X, Roelofs A and Lemhöfer K (2018b) Language selection errors in
speech production. Behavioral and Brain Sciences 22, 1–38. https://doi. switching: language priming or cognitive control? Language, Cognition and
org/10.1017/S0140525X99001776 Neuroscience 33, 139–147. https://doi.org/10.1080/23273798.2017.1363401
Downloaded from https://www.cambridge.org/core. IP address: 46.4.80.155, on 17 Feb 2021 at 16:56:12, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
https://doi.org/10.1017/S1366728919000683You can also read
